Browser demo backlog¶
This list tracks the seven demo categories discussed for the browser gallery. These are basic browser-demo categories, not claims of production calibration accuracy or support for every method. A projection playground alone does not complete an algorithm comparison. Implementation and publication are tracked separately.
Remaining basic demo categories: 0. All seven additions are implemented and published. The limited-coverage / outlier / Huber live-fit controls are also published. The Camera–IMU clock-offset experiment is also published. Its rotation × time residual heatmap is also published. The measured high/low camera-rate window comparison is also published (#317). Linked stereo correspondence close-ups are also published (#318). The interactive sparse stereo 3D view is also published (#319). All three photographed pairs are published for KB4, MEI and DS (#320). The optional side-by-side fisheye model comparison is also published (#321). A linked ACFR observation-by-method residual matrix is a separate addition under validation.
| Demo category | Delivered basic demo | Explicit limits |
|---|---|---|
| Real MEI / Double Sphere fisheye | New native fits to measured corners; real capture 000 rectification, matches and depth | KB4 initialization uses all captures; no independent metric depth truth |
| Camera–LiDAR method comparison | Four ACFR native board methods on one split; separate real A2D2 edge/MI evidence | No cross-dataset ranking or independent extrinsic-accuracy claim |
| GNSS alignment | Real QCOM/u-blox/fused tracks and candidate lever-arm geometry | Fusion is not independent truth; sliders do not estimate a lever arm |
| Radar calibration | Real CAN Radar targets, reported relative speed and candidate yaw | Processed targets, not raw Doppler or a native ego-velocity/extrinsic fit |
| Wheels / INS / vehicle frame | Four measured wheel-speed traces and fused camera FRD velocity axes | Camera frame is not asserted equal to vehicle; scale/time/alignment not separately identified |
| Robot hand–eye | Eight native AX=XB methods on measured ETHZ motion blocks, linked closure origins | Closure is consistency, not absolute accuracy; AX=YB is a further extension |
| Live calibration progress | Actual 6-DoF browser point fit, single-step/run/stop/reset and real residual history | Small measured-board example, separate from the full production solver |
The gallery now has eleven demo tabs. Tabs, algorithm families and this backlog's categories are different counts. A saved recording, manual projection or synthetic regression test must not be relabeled as live optimization or independent accuracy evidence.